Neural Networks Method in Real-Time Forecasting of Apalachicola River Flow

Artificial Neural Networks (ANN) is a computational approach emulating the ability of the biological neural network by interconnecting many artificial neurons. ANN can identify and learn correlated patterns between input data sets and corresponding target values. It can also process problems involving highly nonlinear and complex data even if the data are imprecise and noisy. ANN method is ideally suited for the prediction of chaotic time series such as water resources data, which are known to be very complex and often nonlinear. One of the most popular neural networks is the layered feed-forward neural network with a back-propagation (BP) least-mean-square learning algorithm. In the paper, we present an application of neural network method to forecast the real-time flow change in the Apalachicola River. The Apalachicola River is the lower segment of the Apalachicola-Chattahoochee-Flint basin and is formed by the joining of the Flint and Chattahoochee rivers at the Florida border, which is lies in the southeastern United States with a length of 105 miles and a drainage of 17,200 square miles. Forecasting of river flow, with a warning time of a few hours or days, would benefit water resource management activities in the river basin. In this study, we apply neural network methods to forecast of flow in real time sense based on the historic and current observation of flow at Sumatra station on this river. The network is firstly trained, and then used to forecast flows with lead times varying from 1 day to 5 days. The performance of the neural network model is validated by comparison with observation data. Preliminary results have shown that neural networks can be used to reasonably forecast Apalachicola River flow.